Aiming at the problem that the capacity limitation of inverter made the generalization ability of model worse in the generalized load modeling ,the mechanism that how the inverter capacity limitation influenced the generalized load fault response of the grid-connected point was analyzed.The idea of fitting the piecewise function was proposed and applied to the training of the generalized load artificial neural network class model.Namely, fault samples were jointly trained at the same time to learn different segmentation characteristics. Finally, based on the modeling and simulation results, it is proved that the proposed method can improve the generalization ability and stability of the artificial neural network class model in generalized load modeling.